Deferred capabilities

An item becoming necessary for an included outcome needs an explicit change to the feature specification and its acceptance scope.

Need	Why outside current scope	Discovery context	Disposition
ADR for feature-flag management	A Platform-wide approach extends beyond introducing the MCP flag.	Naming the first flag, `entitlement-remote-mcp-org`, and identifying the need for a code-owned catalog.	Prepare a separate future ADR covering flag types and names, ownership, targeting, defaults and lifecycle. Include a defined flag catalog in code, such as an enum, and how releases detect new flags and provision them in WorkOS through supported APIs. Not a prerequisite for MCP; no ADR drafting or implementation in this scope.
Inventory business tools and projection-backed queries	The foundation exposes only two fixed permission tools and performs no business reads or mutations.	Internal inventory and conversational-demo goals.	Separate Product scope, shared-operation architecture, domain authorization and durable provenance before business exposure.
Business-user evaluation of real workflows	Fixed permission tools do not provide real business tasks. The foundation already has a manual client checklist.	Usability of inventory and query work.	Plan business-user evaluation when real inventory or query functionality is introduced.
Browser-resident assistance through WebMCP	The near-term audience starts work from independent agent conversations; browser integration is a separate interface.	Separate browser-assistance proposal and the distinction between remote and in-page assistance.	Future proposal; pending separate Product scope.
Custom Platform screens for MCP activation, connection management or audit browsing	The feature proposal uses operational enablement, existing user permission administration and client/vendor flows. Whether these suffice is still consequential.	WorkOS product guide's administration controls and review of the initial audiences.	Pending Product decision; required screens need scope expansion and mockups.
Per-client read-only versus write delegation	Agreed initial control is the user's MCP permissions across authorized clients.	Consent and broad delegation discussion.	Deferred; no grant service or client allowlist proposed.
Cross-tenant/environment switching within a connection and service-account automation	Initial access is human-delegated and resource-bound.	Comparison of WorkOS workspace/environment controls with the tenant-specific proposal.	Separate authority/product decision required.
Raw source-data queries, calculations or report execution	The foundation performs no business or source-data query. Report execution introduces a separate data and evidence boundary.	Clarifying what the demo can promise from projection-backed queries.	Confirm initial demo boundary; broader capability pending separate specification.
Destructive deletion, credential operations and MCP policy changes through the agent	Sensitive operations are excluded from the foundation scope.	WorkOS product guide's sensitive-operation exclusions and destructive confirmation pattern.	Separate scope and policy review; an agent confirming its own call is not human approval.
Advanced bulk maintenance	The initial tools do not mutate business data, and later single-target operations do not establish safe batching.	Standalone investigation #1046, reviewed during epic retirement preparation.	Preserve the existing investigation; assess exact-target preview/confirmation, authorization, atomicity or partial success, retry, progress and cancellation before any separate proposal.
Server-side question answering and richer retrieval	Useful projection queries do not require hosting a model or generating answers on the server.	Standalone investigation #1046, outside the deletion set.	Compare client-composed answers with a justified server-side workflow; evaluate grounding, references, freshness, permissions, untrusted content and provider data handling. No provider or implementation selected.
Universal schema-defined command generation and automatic graph translation	The foundation needs no semantic-operation framework. Future business tools require a separately reviewed contract before schema-driven generation is considered.	Separate shared-operation proposal and existing future platform direction.	Future architecture work; no implementation commitment here.